Using HALO Agents and Tools in Content

Events, Articles, and Dialog output nodes can include a HALO agent or HALO tool as an output alongside regular text answers. These outputs behave like any other output in AI Cloud — they can be set as the default, and context variables can be applied to them to control when they fire.

A linked HALO profile is required. Agents and tools are drawn from that profile. Use the Open HALO Studio link in the picker to create or manage them without leaving AI Cloud.

Quick reference

Adding an Agent Output

Click Add agent in the output area of an Event, Article, or Dialog node. A picker lists all agents in the linked HALO profile. Select one and confirm.

How agent outputs work

When an interaction matches a content item with a HALO agent output, AI Cloud hands the conversation to that agent. The agent then handles all subsequent user messages — using its tools, reasoning, and any internal handoffs to other HALO agents — until the agent session ends.

The agent session ends when the agent:

• Breaks out — explicitly configured to return control under specific conditions (e.g. the user asks for something outside the agent's scope)

• Completes — finishes its task naturally

• Hits a HALO-side fallback — no agent in the HALO profile can handle the interaction

In all three cases, the agent session closes and the next user message goes back through AI Cloud's full NLP pipeline. If the pipeline routes to HALO again for that message, a new HALO interaction begins from scratch.

For more on how agents work, including breakout configuration, see Agents in the HALO documentation.

Adding a Tool Output

Click Add tool in the output area. The picker lists all tools in the linked HALO profile, each with a short description. Select a tool and click through to the configuration step.

Mapping input parameters

Each tool defines its own input parameters. Required parameters are marked with an asterisk. For each parameter, either select a conversation variable (whose current value will be passed in) or type a static value directly.

Storing tool output

Tool output can optionally be stored in a conversation variable for use later in the conversation. Add one or more output mappings using the + Add button. Each mapping has:

• Type — Payload: stores the tool's full JSON response in the selected conversation variable.

• Type — Property: extracts a specific field from the response using a JMESPath expression (entered in the Property field), and stores that value in the selected conversation variable. Use this when you only need a single field from a larger response object.

The Variable dropdown shows all conversation variables defined in your project. If you need to surface the output in a response, reference that variable in a subsequent answer.

How tool outputs work

When the content item fires, AI Cloud invokes the tool with the mapped inputs. The tool executes once, returns its output, and any configured variable mappings are applied. The next user message then returns to AI Cloud's NLP pipeline — tool outputs are single-shot and do not hold the conversation the way agents do.

For more on how tools work, including step configuration and output handling, see Tools in the HALO documentation.

Context and Default Behaviour

Agent and tool outputs follow the same rules as text answers:

• One output per content item can be set as the default — fired when no context matches.

• Context variables can be applied to agent/tool outputs to control when they fire relative to other outputs on the same item.

• Multiple outputs (text, agent, tool) can coexist on the same content item, each with different context conditions.